ITPEC FE Morning April 2020 Question 4

Training on a dataset with answer keys, then answering new inputs — the known answers act as the supervisor.

In supervised learning each training example is labelled: an input and the output it should produce. The model is adjusted to reduce the gap between its predictions and those labels, so it generalises to new data it has not seen. Classification and regression are the typical tasks.

Answer (c)

Why not others:

• •(a) — finding unknown patterns in data without labels is unsupervised learning, such as clustering
• •(b) — finding the best action by evaluating the results of possible actions is reinforcement learning, which learns from rewards rather than answer keys
• •(d) — predicting a user's actions from collected behaviour data is an application, such as recommendation, and says nothing about training on labelled answers

Key rule: Supervised learning has labelled answers, unsupervised learning has none, and reinforcement learning has only a reward signal for its actions.

Basic Theory · AI-written, may contain errors